Full course coming soon

AI System Performance: Evaluating Hardware, Memory, and Throughput

7 chapters · 32 lessons

1. 1. System-Level Performance Fundamentals

   4 lessons

   Understand why AI performance is a system property and learn the factory model for evaluating throughput.

2. 2. Accelerator Architecture and Compute Units

   4 lessons

   Learn how specialized processors achieve parallelism and where their architectural choices create tradeoffs.

3. 3. Memory Hierarchy and Bandwidth Bottlenecks

   5 lessons

   Diagnose when memory, not compute, limits performance and learn strategies to maximize data movement efficiency.

4. 4. Interconnects and Multi-Accelerator Systems

   5 lessons

   Understand how accelerators communicate and when interconnect topology becomes the limiting factor.

5. 5. Software Stack and Orchestration

   5 lessons

   Learn how compilers, runtimes, and schedulers translate model code into efficient hardware execution.

6. 6. Training Workload Optimization

   4 lessons

   Apply system thinking to large-scale training, balancing synchronization overhead with compute efficiency.

7. 7. Inference Deployment and Cost Optimization

   5 lessons

   Design inference systems that meet latency and throughput requirements within budget and power constraints.